의사를 위한 ChatGPT 활용 사례 (2025.2)

English summary

This video presents practical demonstrations of ChatGPT-4o, Canvas, Custom GPTs, and DALL-E for clinical education, patient education, research organization, and image production. The main demonstrations cover comparative analysis of tuberculosis guidelines, creation of latent tuberculosis patient education materials and FAQs, configuration of Custom GPTs for repetitive tasks, graduated explanations of medical concepts, and image generation with region editing.

Key points

• ChatGPT-4o was used to compare PDF versions 4 and 5 of the Korean tuberculosis guidelines, including changes in diagnostic criteria, drug-resistant tuberculosis regimens, and latent tuberculosis infection guidance. The demonstration also examined detailed changes for clofazimine, such as indications, dosing regimens, and precautions.
• The uploaded guideline content was used to draft patient education material for isoniazid monotherapy and to generate 10 frequent patient questions covering transmission risk, the need for treatment, treatment duration, missed doses, dietary restrictions, and liver function monitoring.
• Canvas was used to write React code for a latent tuberculosis FAQ accordion application, with color and design changes made through natural language prompts. The presentation also described sharing the generated tool and allowing account holders to modify its code. These demonstrations did not provide validation of deployment safety or performance in real clinical settings.

Audience

Physicians, clinical specialists, resident physicians, medical educators, and healthcare researchers considering generative AI for clinical work and education

Scope and limitations

This summary is based only on the supplied evidence notes from the video. The presentation is demonstration-oriented and does not report systematic evaluations or comparative study results for output accuracy, patient safety, privacy protection, clinical effectiveness, or long-term workflow impact. Document analysis requires a model that supports file attachments, and generating code in Canvas may take noticeable time. Image-generation models may struggle with exact quantities and spatial placement, while complex edits may lack sufficient fidelity and require manual post-processing.

Research context

The material is thematically connected to Sangzin Ahn's research and educational portfolio through its focus on applied medical AI, document-based knowledge extraction, generative AI for clinical education, and medical data science tools.